Using microbial fuel cell output metrics and nonlinear modeling techniques for smart biosensing.

Microbial fuel cells (MFCs) are promising tools for water quality monitoring but the response peaks have not been characterized and the data processing methods require improvement. In this study MFC-based biosensing was integrated with two nonlinear programming methods, artificial neural networks (ANN) and time series analysis (TSA). During laboratory testing, the MFCs generated well-organized normally-distributed peaks when the influent chemical oxygen demand (COD) was 150 mg/L or less, and multi-peak signals when the influent COD was 200 mg/L. The area under the response peak correlated well with the influent COD concentration. During field testing, we observed normally-distributed and multi-peak profiles at low COD concentrations. The ANN predicted the COD concentration without error with just one layer of hidden neurons, and the TSA model predicted the temporal trends present in properly functioning MFCs and in a device that was gradually failing. This report is the first to integrate ANN and TSA with MFC-based biosensing.